Paper Overview
Research Field: NLP / Image Generation Authors: Wei Zhou, Xiongwei Zhu, Zelin Xu Published: 2026-06-27 arXiv: 2606.27377
Abstract
Modern image generation demands a single model that unifies diverse capabilities, including text-to-image (T2I), local editing, and global editing. However, these capabilities are rarely naturally aligned and often conflict. For instance, editing tends to degrade T2I performance, while global and local editing interfere with each other. Consequently, effectively composing these capabilities has become a central challenge for image generation model training.
To tackle this, the authors introduce DanceOPD, an on-policy generative field distillation framework. The key idea is that each capability should learn from the model's own generated distribution (on-policy), rather than from misaligned or off-policy data, thereby avoiding misalignment and conflicts between capabilities.
Key Points
- Problem: T2I generation, local editing, and global editing conflict when trained together in a unified model.
- Approach: On-policy generative field distillation — each capability learns from the distribution the model itself generates.
- Goal: A single unified model that composes multiple image generation capabilities without mutual degradation.
*Auto-collected on 2026-06-27.*